The Extreme Control Choice

Transcription

1 The Extreme Control Choice Preliminary and Incomplete Ronald Anderson Ezgi Ottolenghi David Reeb* May 14, 2017 Abstract Firms with limited voting shares, dual class firms, persist over time despite the widespread view of a corrupt governance system. We find that founders and their heirs control 89% of existing dual class firms, indicating family control and voting rights represent important covariates. Our initial analysis indicates investors place significant discounts on dual class. However, after controlling for family ownership, the discount disappears. Without family owners, dual class firms exhibit a 19% premium relative to single class firms. The study further shows that dual class firms cluster in a small number of industries with high brand maintenance, high intangible asset levels, and high product market power. Strikingly, we find that shareholders of dual-class firms earn excess returns of 300 basis points per year relative to single class firms. Overall, our study does not provide evidence that dual class structures negatively affect outside shareholders; rather, these structures appear to occur with specific ownership and industry characteristics. JEL classification: G31; K22; L22; G23 Keywords: Super voting rights, dual class shares, product quality, short-termism, family firms * Anderson ( ron.anderson@temple.edu) is with Temple University, Ottolenghi (ezgi.ottolenghi@temple.edu) is with Temple University /Texas Tech and Reeb (dmreeb@nus.edu.sg) is with National University of Singapore.

2 The Extreme Control Choice Firms issue a variety of distinctive securities or claims on their assets, creating conflicts of interest among stakeholders (Aghion and Bolton, 1992). Issuing common stock with limited voting rights, commonly known as dual class shares, represents a notorious example of multiple equity classes that the media routinely criticizes. The academic and business press describe limited voting shares as severely harming outside investors, highlighting that insiders with superior voting rights enjoy substantial firm control with limited economic exposure. 1 Masulis et al. (2009) indicate that these additional control rights lead to value destroying acquisitions and allow managers to extract private benefits from the firm. Zingales (1995) documents that the different share classes trade at different prices even though both classes maintain equal cash flow rights. Gompers et al. (2010) report that after controlling for firm and industry characteristics, multiple share class firms exhibit lower valuations relative to single class firms. In short, a prominent academic literature documents that issuing limited voting shares produces substantial conflicts of interest among firm shareholders and harms outside equity claimants. Evolutionary economic arguments indicate that governance mechanisms that aim to exploit investors should die out over time (Nelson, 1995). Kole and Lehn (1997) observe that firms with suboptimal governance structures face extinction, suggesting that mechanisms surviving the natural selection process may be efficient outcomes. The hypothesis that dual class shares embody poor corporate governance implies that these firms should be short lived and their use should decline through time. Surprisingly, dual class structures persist over time and comprise almost one in 11 listed firms in the U.S. (Hong, 2013). Figure 1 shows that instead of disappearing, dual class firms remain a popular and growing choice with initial public offerings. The common disparagement from business and academic reports surrounding dual class shares raises the question of Why do firms establish these structures? We begin by analyzing the owners or originators of these multiple equity class firms. Focusing on the industrial firms (i.e., 1 A Forbes (2012) article suggests that dual class share structures benefit corporate insiders at a severe expense to outside shareholders. A Bloomberg report, in 2013, indicated that dual class share firms exhibit weak internal controls and experience more external conflicts, concluding Buyers Beware. Calpers, in 2011, announced that dual class shares are a corruption of the governance system, while the Wall Street Journal declared the following year that super voting shares undercut outside shareholders. 1

3 excluding utilities and financial firms) of the Russell 3000, we collect information on ownership composition, differential voting rights, and capital structure. Our sample comprises 2,379 firms or 24,724 firm-year observations spanning from 2001 through Dual class firms comprise 9.4% (2,333 firm-year observations) of the Russell 3000 industrial firms. Notably, founders or their heirs (controlling- or family- shareholder firms) constitute nearly 89% of dual class firms. The remaining 11% of dual class firms fall within two categories. First, about 7% arise in firms with diversified shareholder bases and tend to have brief or short-lived lives after going public (e.g., Mondelez International). Second, the remaining 4% represent legacy structures where the founders initiated multiple security classes, exited the firm, and the firm continues with dual class shares (e.g., Hershey Corporation). 2 Our analysis points to several immediate differences between dual class family firms and other firm types. Strikingly, dual class family firms tend to be substantially larger, older and exhibit superior operating performance than single class firms (family or nonfamily). Family owners in dual class firms hold substantially greater cash flow rights and voting rights than family owners in single class firms. Dual class family owners hold, on average, about 31.3% of their firm s cash flow rights and 58% of voting power. Single class family shareholders in contrast, retain 22.9% of the cash flow and voting rights. Controlling shareholders in dual class share firms possess significant economic exposure to the firm, which they couple with dual class structures to retain formal control of firm decision-making (Aghion and Tirole, 1997). To assess investor perceptions of dual and single class firms, we use Tobin s Q as a proxy for firm valuation. Because of the nearly inseparable link between family ownership and dual class share structures, we divide our sample into four mutually exclusive groups: dual class family firms (8.4% of observations); single class family firms (25.6%); dual class nonfamily firms (1.1%) and; single class nonfamily firms (65.0%). Relative to a benchmark of single class nonfamily firms, we find that dual class family firms exhibit valuation discounts of 11.8%, followed by single class family firms with a discount of 3.7%. Notably, dual class nonfamily firms, although a minute subset of all firms (about 1.1%) experience valuation premiums of 20.8%. Using a matched sample rather 2 We also investigate firm equity ownership structure at the time of IPO. Using a matched sample of dual-class and single class firms, we find that family (nonfamily) firms comprises 58.8% (41.2%) of dual class firms at IPO. Within the single class firms at IPO, we find that family (nonfamily) firms constitute 44.6% and 55.4% of firms. 2

4 than the Russell 3000 industrials, we find similar results. Outside investors appear to place discounts on the family firm organizational structure with particularly deep discounts on the family-firm, dual-class combination. Yet, once dual class structures become separate from family ownership, outside investors tend not to discount this organizational form. Early literature highlights that dual class structures protect investment opportunities from hostile takeovers and potentially encourage entrepreneurs and family owners to invest in organization specific capital whose returns may otherwise be appropriable by outsiders (Williamson, 1975; Klein, Crawford, and Alchian, 1978; DeAngelo and DeAngelo, 1985). More recently, Banerjee and Masulis (2016) argue that dual class shares allow insiders to garner private benefits while still raising funds to invest in positive NPV projects. To investigate the argument that super voting shares protect firm investment opportunities, we examine dual class structures relative to industrial structure. Casual observation indicates dual class firms cluster across a small number of industries. Specifically, over 53% of all dual class firms reside in just seven of the 48 Fama-French industry groups, e.g., communications (19.0%), retail (8.5%), business services (6.7%), print and publishing (5.3%), electronic equipment (5.0%), apparel (4.5%), and food products (4.2%). Firms within these industries often maintain high visibility with customers and the consuming public, suggesting that these industries offer private benefits of control or alternatively, may require high levels of monitoring to ensure product and service integrity (Gompers, 1995; Giroud and Mueller, 2010). Internet appendix I provides illustrative examples of dual class share firms in each of these seven industries. No dual class firms exist in eight of the Fama-French industry groups that the media often characterizes as traditional or old-line industries (i.e., tobacco, shipbuilding, railroad, etc.). 3 The concentration of dual class firms across a small number of industry groups suggests that industrial structure and/or opportunities play an important role in family owners decision in establishing firm control rights. To gain further insights into market forces influencing the dual class choice, we examine equity ownership structures relative to industry traits or characteristics. 4 We investigate whether family owners appear more likely to employ dual class structures (relative to exiting or single class 3 These industry groups are; tobacco products, fabricated products, shipbuilding and railroad equipment, defense, precious metals, coal, real estate, and almost nothing. 4 Appendix A provides a discussion of family owners equity structure choice when making the choice to go public. 3

5 structures) when firms face: (i) high growth opportunity potentials; (ii) long-term investment environments (Aaker and Jacobson, 1994) and; (iii) high levels of firm specific investments assets (e.g., product quality or image, product market power, and intangible assets (Phillips et al., 1983; Kogut, In our analysis, we do not preclude that family owners use dual class structures to exploit minority shareholders, which could occur in a variety of industrial settings. Rather, our analysis seeks to evaluate industrial and competitive pressures associated with dual class share structures. We find the dual class family firms are only 40% as likely to exist in high growth industries relative to single class nonfamily firms (base case). Using industry growth rates or innovation, our results suggest that dual class family firms appear to shun high growth industries. Similarly, we find little evidence to suggest that dual class family firms arise in industries with long investment horizons. Our analysis however, indicates that dual class structures are nearly four times as likely to reside in industries requiring high branding levels and over three times more likely to reside in industries with high levels of intangible assets relative to single class nonfamily firms. Dual class firms are also nearly 2.7 times more likely to exist in industries with high levels of product market power (Karuna, 2007). Our results generally indicate that family owners do not use dual class shares to protect future growth opportunities but instead imply a mechanism that protects firmspecific investments such as branding and product market power from outside investors and/or interference. The analysis thus far, suggests that outside investors, with little influence in firm decisionmaking, purchase shares in dual class firms at significant discounts relative to their single class family- and nonfamily- peers. Further, we document a clustering of dual class structures in industries with high levels of branding, intangible assets, and product market power; suggesting an extreme governance choice to protect firm specific investments. Given the large discounts when buying shares, the question arises as to whether outside shareholders fare any worse for investing in dual class firms than their single class counterparts. We investigate this question by examining stock price returns. Our stock return analysis presents a striking contrast to the valuation results. Dual class family firms significantly and economically outperform their counterparts. Using industry adjusted 4

6 returns, market adjusted returns, and size and book-to-market adjusted returns, the results indicate that a simple that a buy-and-hold strategy of dual class family firms earns excess returns of about 350 basis points per year relative to our benchmark (single class nonfamily firms). The analysis allows a rank ordering of stock price returns based on firm organizational structure. Dual class family firms provide the highest returns followed by single class family firms, with nonfamily firms single and dual providing similar returns but at the bottom of the ranking. In our matched sample analysis, we confirm our return results by finding that dual class family firms earn excess of about 500 basis points per year over that of benchmark group. The results of our stock return analysis arguably indicate that shareholders earn a risk premium for investing in dual-class family firms relative to other organization structures. In our returns analysis, after controlling for market, industry, time, and firm specific factors (size, age, leverage, operating performance, risk, and growth opportunities), we continue to observe that dual-class family firms earn excess stock returns of about 350 basis points per year. One potential explanation for the excess return centers on outside investors earning a premium for bearing family expropriation risk, with the risk being particularly acute in dual class firms. To explore the expropriation risk argument, we delineate family firms based on whether founders, descendants, or professional managers hold power and influence. Academic studies often advocate that the benefits to family ownership, if any, stem primarily from founder and professional-manager control (Fahlenbrach, 2009). Villalonga and Amit (2006) argue that descendant control often leads to poor corporate execution and financial performance because of weak ability and work ethic. The literature on family firm control thus suggests that family expropriation may be particularly acute when descendants control the firm versus their gifted parent or professional managers. Of course, both single- and dual- class family owners can extract private benefits from the firm. We argue however, that outside investors of dual class family firms with little to no voice in firm decision-making bear a particularly high expropriation risk relative to single class family-firm investors or nonfamily firm investors. The valuation analysis suggests a rank ordering across family firm type. Dual class descendant firms bear the deepest discounts (-19.7%), followed by single class descendant firms (-14.5%), then founder and professionally managed dual class firms (both at -8.3%), and finally single class 5

7 founder and professional firms with no valuations discounts, all relative to our benchmark of single class nonfamily firms. The returns analysis indicates that all three subcategories of dual-class family firms earn superior excess returns versus single class nonfamily firms. Founder, descendant, and professionally managed dual-class firms earn excess industry returns of 280, 410, 420 basis points, respectively, more per year than the benchmark (single class nonfamily firms). Notably, within the single class family firms, we find that descendant firms earn excess returns of about 520 basis points per year versus the benchmark. Neither single class founder nor professionally managed firms outperform the benchmark. Matched sample analysis provides similar results. Our family firm type analysis provides two important inferences. First, although we find valuation discounts across all types of dual class family firms, descendant firms appear to suffer from the largest discounts, about 19.7% and nearly double the discounts of founder or professionally managed firms. The analysis suggests that outside investors place relatively severe discounts on descendant firms, regardless of share structure. Second, after controlling for time, industry, and host of firm specific factors, we find that outside investors earn superior buy-andhold returns across all subcategories of dual class family firms about 400 basis points. The results further indicate that investors earn excess returns in single class descendant firms of about 520 basis points. The return analysis thus, points to investors earning a risk premium for holding dual class family firms and for holding single class descendant firms. Our study makes several important contributions to the literature. First, we find that nearly all existing dual class structures arise from entrepreneur and family ownership. Across the Russell 3000 industrial firms from 2001 through 2015, we show that 93% of dual class firms either still have the founding family (89%) as owners or the founding family initiated the structure and then exited (4%). Dual class family firms are significantly larger than are their single class counterparts, appearing similar in size to nonfamily firms. Moreover, these controlling shareholders hold substantially greater economic exposure in dual class firms relative to single class firms. Second, both single and dual class family firms exhibit large valuation discounts relative to nonfamily firms. In contrast, among nonfamily firms, dual class share firms exhibit a 20% premium relative to their single class peers. Because family owners establish the dual class structure and sell their shares into the market at a substantial discount, these owners appear to bear the cost arising from their organizational structure. 6

8 Third, our analysis highlights that dual class share structures yield both costs and benefits. The observable costs appear to arise from valuation discounts on dual class family firms. The benefit arises from the superior buy-and-hold returns that outside (and family) investors accrue from holding shares in dual class family firms. We find excess returns of about 350 basis points per year for holding dual class family shares relative to single class nonfamily shares, suggesting outside investors require a family firm risk premium for holding these shares. Fourth, our evidence suggests that dual class organizations do not randomly arise at the whim of corporate insiders but rather appear in industries with high levels of branding, high levels of intangible assets, and industries with substantial product market power; suggesting that firmspecific capital investment influences the dual class choice (Tirole, 2005). Dual class structures allow the founding owners an opportunity to take their investment to market, raise capital, but still protect the investment from outside interference. In contrast, to conventional wisdom in the governance literature that dual class structures exploit outside investors and destroy firm value, our analysis implies that family owners use super voting shares to protect firm specific investments and that outside equity holders earn superior returns on their investment. The remainder of the paper is organized as follows. Section I provides a summary of the data and descriptive statistics. Section II presents the empirical results. In Section III, we conclude. A. Sample I. Data and Descriptive Statistics For our empirical analysis, we start with the Russell 3000 firms as of December 31, We exclude regulated public utilities (SIC codes 4812, 4813, 4911 through 4991) and financial firms (SIC codes 6020 through 6799) because government regulation potentially affects firm equity ownership structure. Data on equity ownership structure (i.e., single- and dual- class), inside owners cash flow and voting rights, and the family s role in management comes from annual corporate proxy statements. We gather firm specific control and primary variables from CompuStat and stock return information from the Center on Research in Security Prices (CRSP). To control for survivorship basis, we allow firms to exit and re-enter the sample. Our final sample consists of 2,379 industrial firms (non-financial and non-utility) or 24,724 firm-year observations, spanning from 2001 through Notably, as our base sample starts in 2001, our data does not 7

9 include many of the recent technology firms such as Alphabet, Facebook, or Alibaba that went public as dual class shares and experienced impressive and sustained stock returns. B. Equity Ownership Structure Firms issue one or multiple classes of common equity. We define single class firms as those where the firm establishes one class of common equity that grants shareholders equal cash flow and voting rights on a per share basis. Dual class firms are those issuing two or more outstanding classes of common stock. The class with the largest number of shares outstanding typically receives equivalent cash flow and voting rights, e.g., one share, one vote, one dividend. The class with the smaller number of outstanding shares most frequently receives 10-votes per share and one cash flow right per share, e.g., one share, 10-votes, one dividend. Although we find the most prevalent voting-right differential to be 10-to-1 between the two classes, for some firms, we observe that the super voting class provides 100 or more votes to their holders. Our analysis uses a binary variable that equals one for dual-class firms and zero otherwise (i.e., single class firms). Family firms are those where the family (founders and/or their descendants) continues to maintain a 5% or larger voting stake in the firm. Notably, Shleifer and Vishny (1986) and Villalonga and Amit (2006) use a definition of 5% or more of the cash flow rights. For the single class firms in our sample (90.6%), using cash flow or voting rights yields the same level of family influence because one share provides one vote and one cash flow right. For dual class firms (9.4%) however, we note that family owners, on average, hold 30.2% of the cash flow rights and 58.2% of the voting rights. To ascertain the effect of family influence on firm characteristics, we focus our analysis on voting power. To be classified as a family firm, a family member does not necessarily need to hold the COB, CEO or director position. The classification refers to families maintaining a minimum voting stake of 5%. The initial analyses use a binary variable that equals to one when families hold a 5% or larger voting right in the firm and zero otherwise. In subsequent analysis, we also use a continuous measure of family ownership and voting power. Firms through their public filings frequently do not provide information on whether founding-family members retain equity stakes or hold managerial posts and director seats. Although regulations stipulate that firms disclose any shareholder with a 5% stake or larger equity stake, firms do not typically disclose if the shareholder 8

10 is part of the original founding family. To ascertain founders and their subsequent lineage and involvement in the firm, we examine corporate histories for each of the 2,379 firms in our sample. Corporate histories come from ReferenceforBusiness.com, FundingUniverse.com, Gale Business Resources, and from individual companies. C. Valuation and Performance Measures To assess outside investors perception of firm value, we develop a proxy for Tobin s Q by using the ratio of the market value of total assets to the book value of assets (Masulis et al., 2009; Fahlenbrach, 2009). The market value of total assets is the sum of the book value of assets and the market value of common stock less the book value of common stock. We measure the market value of common equity at the end of each calendar year. Notably, for dual class firms, we measure the market value of common equity as the sum of the two outstanding classes multiplied by the share price of each class. If one of the classes does not publicly trade (i.e., entirely owned by corporate insiders), then we use the share price of the traded class. By using the share price of the traded class, we likely understate the market value of dual class firms as the superior voting class typically trades at a premium to the other voting class. We measure firm performance using three measures of stock price returns. First, we industryadjusted returns that equal each firm s annual return less the annual return of the corresponding return of the Fama-French (1997) industry code (based on the 48 industries). Second, we calculate market adjusted returns as the firm s annual return less the return on the CRSP value-weighted market return (De Bondt and Thaler, 1985). Third, we use size and book-to-market adjusted returns that equal each firm s annual stock return less the Fama-French size and book-to-market benchmark portfolios (Faulkender and Wang, 2006). D. Control Variable Measurement Previous literature indicates that dual class firm performance varies with firm characteristics. We measure firm size as the natural logarithm of total assets at fiscal year-end. Firm age is the natural log of the number of years since the firm s inception and captures firm and industry maturity as well as the family s investment period with the firm. We use return on assets to control for operating performance and measure it as operating income before depreciation scaled by total assets. Firm risk is the standard deviation of stock returns for the previous 36-months. Because 9

11 family shareholders may be more reluctant to use debt in the firm s capital structure (Anderson and Reeb, 2004) and because debt exhibits a strong relation to Tobin s Q, we control for leverage with the ratio of long-term debt to total assets. We control for firm growth opportunities with the ratio of R&D expense to sales. In the multivariate analyses, we control for industry effects with the Fama-French 48-industry codes and for time effects with year binary variables. E. Matched Sample Dual class firms comprise less than 10% of our firm-year observations. To assess the robustness of our results with better comparability in sample size between our four firm categories and to control for other firm characteristics potentially influencing equity structure, we develop a matched sample using coarsened exact matching (CEM) (Iacus, King, and Porro, 2009). The matching criteria are exact Fama-French industry code, total assets and firm age. We conduct three matches. These are dual class family firms to; (i) single class nonfamily firms, (ii) single class family firms, and (iii) dual class nonfamily firms. The first match between dual class family firms and single class nonfamily results in 2,868 firm-year observations evenly split between the two firm categories. The second match between dual class family firms and single class family provides 2,278 firm-year observations evenly split between the two groups. The third match between dual class family firms and dual class nonfamily firms yields 288 firm-year observations, again, evenly split between the two firm groups. We then combine the three matches in a single sample that yields a total of 4,258 firm-year observations. The final, combined matched sample consists of 1,541 dual class family firm observations (36.19%), 1,139 single class family firm observations (26.75%), 1,434 single class nonfamily firm observations (33.68%), and 144 dual class nonfamily firm observations (3.38%). 5 Table 2, Panel A, columns 5 through 8 present the summary statistics for the matched sample and indicate a relatively homogeneous match between single and dual class firms. Mean tests indicate no difference in firm size, firm age, firm risk, leverage, return on assets, and R&D to sales for the matched single and dual class firms. 5 Assuming a unique match for each dual class family to each of the three different firm categories (single class family firms, single class nonfamily firms, and dual class nonfamily firms), our final would consist of 2,717 dual class family firms. Our final sample however, consists of 1,541 dual class family firms, indicating that 1,176 dual class family firms match in two or more of the other firm categories. 10

12 II. Univariate Analyses A. Descriptive Statistics Table 2, Panel A provides summary statistics for the full (matched) sample of 24,724 (4,258) firm-year observations. We show mean values for the full sample (column 1), dual class firms (column 2), single class firms (column 3) and also t-values for difference of mean tests between single- and dual- class firms (column 4). Across the Russell 3000 industrials from 2001 through 2015, dual class firms and single class firms constitute 9.44% (2,333 observations) and 90.56% (22,391 observations), respectively, of the sample. Dual class firms exhibit substantial differences from single class firms. Notably, we observe that dual class firms are larger (total assets: $5,553 billion versus $4,918 billion), older (53.5 versus 45.1 years) and substantially more levered (25.1% versus 19.3% of total assets) than their single class counterparts. These firms also exhibit better operating performance and stock return performance their single class firms. On average, dual class firms exhibit operating performance (ROA) of 10.33% per year while single class firms performance comes in at 8.48%. Using industry adjusted returns and market adjusted returns, we find the dual class stocks outperform single class firms by over 200 basis points per years. In stark contrast to the operating and stock return results, we find the dual class firms exhibit steep valuation discounts relative to single class firms. Specifically, our univariate results indicate, on average, Tobin Q s of 1.71 and 1.99 respectively, for dual and single class firms. The differences between dual and single class firms for size, age, debt levels, operating performance, stock returns, and Tobin s Q are significant at the 5% level or better. Our univariate analysis suggests that dual class firms tend to be larger, older firms with superior accounting and market performance although suffering significant valuation discounts relative to their single class counterparts. Dual class firms appear to be a manifestation of founders and their families. Table 2, Panel B provides summary statistics for single and dual class firms segregated into family and nonfamily firms. Within the dual class set, family firms comprise 88.7% (2,070) of firms with the remaining 11.3% (263 observations) falling under a nonfamily categorization. In further analysis, we examine corporate histories for the 264 firm-year observations that comprise the dual class nonfamily firms. A substantial number of these observations (99 observations, 38%) are originally family firms where the family owners exited their equity stake and the firm continues to operate with the dual 11

13 class structure. For instance, Milton Hershey established the firm bearing his name in the late 1800 s. With no heirs to leave his fortune, in 1909, he bestowed ownership of the firm to the Hershey School Trust that controls the firm (76% of votes) through the super-voting B shares. The remaining dual class nonfamily firms (165 observations, 62%) appear to arise from special corporate transactions such as Cooper Industries, Inc. where one of their subsidiaries holds the entirety of the B shares, which have no voting rights, thereby preventing voting-power dilution of class A-shareholders. The univariate analysis clearly points to dual class firms originating from founders and their families. The analysis also points to significant differences between dual class family firms and single class family firms. Dual class family firms are significantly older (53.9 vs years), larger ($5,554 vs. 2,751), less risky (13.83% vs %), and use substantially more debt (24.9% vs. 16.0%) than single class family firms. We also find that dual class family firms exhibit significantly better operating performance relative to single class family firms (10.84% vs. 8.25%). Our descriptive statistics show that family shareholders in dual class firms own 31.3% of the firm s cash flow rights and control 58.1% of firm voting power. Single class family owners in contrast, hold 22.9% of the cash flow and voting rights. Figures 2 and 3 show the distribution of family cash flow and voting rights for dual class and single class firms. Although voting control of dual class family owners outstrips their economic interests, our analysis does not indicate that these influential owners hold small equity stakes. Rather, we find that dual class family owners hold significantly larger equity stakes than their single class counterparts. Table 3 segregates dual class and single class firms by the Fama-French industry codes. We observe a clustering of dual class firms in a small number of industries. Specifically, nearly 38% of dual-class firms reside in just four of the 48 Fama-French industry codes. The communications industry (FF48=32) accounts for over 19% of dual-class firm observations, followed by retail (FF48=42) with 8.5%, print and publishing (FF48=8) with 5.3%, and electronic equipment (FF48=36) with 5.0% of dual class observations. Ten industry groups account for 64.4% of dual class firm observations. 6 Another ten industries that the media often characterizes as old line (i.e., 6 These are; communications (FF48=32), retail (FF48=42), print and publishing (F48=8), electronic equipment (FF48=36), apparel (FF48=10), business services (FF48=34), entertainment (FF48=7), wholesale (FF48=41), food products (FF48=2), and transportation (FF48=40). 12

14 railroad, tobacco) contain no dual class firms. 7 The clustering of dual class firms suggests that industry characteristics appear to play an important role in entrepreneurs and family owners decision to establish and maintain dual class structures. Overall, our univariate analysis indicates dual class firms tend to be larger, older, and less risky; use substantially more debt, and exhibit superior operating and stock return performance relative to their single class peers. The results further indicate that nearly all dual class structures arise from founder and family ownership. Finally, we find a clustering of dual class firms among a small number of industries, suggesting that industrial structure arguably plays into the dual class choice. III. Dual Class and Family Ownership The univariate analysis indicates that dual class structures and family ownership represent important covariates that is, dual class structures tend to simultaneously exist with family ownership. Prior literature indicates that firms issuing two classes of equity securities suffer from governance problems that negatively affect firm valuation and performance (Masulis et al., 2009; Gompers et al., 2010). Extant research however, remains relatively silent on the relation between family ownership and dual class shares and their effects, if any, on firm performance and valuation. We begin our multivariate analysis by examining the relation between family influence and dual class shares on firm valuation. To examine the association, we use the following specification; Tobin s Q it =α + β 1(Dual Class) + β 2(Family Firm) + β 3(Dual Class * Family Firm) + β XX + ε t (1) Table 1 provides the variable definitions. X represents a vector of control variables that include natural log of total assets, natural log of firm age, leverage, return on assets, firm risk, and R&D expense to sales. The analysis uses binary variables to capture dual class firms and family firms. The reference variable for the regression specification is single class firms. We control for serial correlation and heteroskedasticity using the Huber-White sandwich estimator (clustered on firmlevel identifier) for the standard errors on the coefficient estimates. Table 4 presents the regression results. Consistent with prior research, we find that dual class firms exhibit significant valuation discounts relative to single class firms. The coefficient 7 These are; tobacco products (FF48=5), fabricated products (FF48=20), shipbuilding and railroad equipment (FF48=25), defense (FF48=26), precious metals (FF48=27), coal (FF48=29), utilities (FF48=31), banking (FF48=44), real estate (FF48=46), and almost nothing (FF48=48). 13

15 estimate on dual class in column 1 indicates that firms with dual-class structures experience valuation discounts of 6.4% versus single class firms. We calculate this percent difference as the coefficient estimate on dual class divided by average Tobin s Q value for the full sample (= /1.96 = 6.43%). The regression specification in column 2 includes the family firm variable along with the dual class variable. Notably, after controlling for family presence, we no longer find a significant relation between Tobin s Q and dual class firms. The family firm variable however, bears a negative and significant relation to firm performance; suggesting that family firms exhibit valuation discounts of about 5.1% relative to the reference or benchmark variable (single class nonfamily firms). 8 Column 3 presents a regression specification with variables for dual class, family firm, and an interaction term between dual class and family firm. Although not significant at conventional levels, we find that the standalone dual class variable exhibits a positive relation to firm performance. That is, after controlling for family presence and the interaction of dual class and family presence, the coefficient on dual class changes from a negative to positive estimate. The standalone family firm continues to bear a negative relation (at the 10% level) to firm performance. Finally, we find a large, negative coefficient estimate on the interaction term between dual class and family firm; suggesting that firm valuations particularly suffer from the combination of family ownership and dual class structures. In an F-test examining whether Tobin s Q differs between dual class family firms versus single class nonfamily firms (β 1+β 2+β 3=0), we reject the null (F=12.07, p=0.001) and conclude that dual class family firms exhibit significant firm valuation discounts relative to single class nonfamily firms. Our analysis with the interaction term (column 3) suffers from a relatively severe multicollinearity problem because nearly all dual class firms are also family firms. We find a correlation coefficient of 93.65% between the standalone dual class variable (β 1) and the interaction term between dual class and family firm (β 3). Consequently, the coefficient estimates on the variables of interest in the specification with an interaction term are relatively unstable. 9 To 8 We calculate this as the coefficient estimate on family firm divided by the average Tobin s Q for the full sample; /1.96 = -5.05%. 9 The variance inflation factors on dual class and the interaction of dual class and family firm are 6.52 and 6.90, respectively. 14

16 mitigate the multicollinearity issue, we segregate our sample firms into four mutually exclusive firm groups; dual class family firms (8.4% of observations), dual class nonfamily firms (1.1% of observations), single class family firms (25.6% of observations), and single class nonfamily firms (65.0% of observations). Column 4 of Table 4 presents an alternative specification that delineates the four firm types relative to Tobin s Q with single class nonfamily firms as the reference (omitted) variable in the regression. Three notable points from this specification. First, dual class nonfamily firms exhibit a marginally positive relation to firm performance; suggesting that a dual class structure in and of itself does not appear to be a detrimental factor in influencing firm valuations. Second, family firms with a single class of common equity exhibit a negative relation to Tobin s Q (p-value=11%), indicating that outside investors marginally discount the family organizational form. Third, consistent with our earlier results, we find that firm valuation particularly suffers under the family firm-dual class combination. The coefficient estimate on dual class family firm suggests that this organizational form experiences an 11.8% discount relative to single class nonfamily firms. 10 Our earlier univariate analysis documents that dual class firms tend to cluster in a small number of industries, tend to be larger, and tend to be older than single class firms; suggesting that industrial or firm characteristics potentially influence the valuation results. To investigate this possibility, we repeat our analysis using the matched sample outlined in Section I.E. Table 4, column 5 shows the multivariate analysis for the matched sample. Again, we find little evidence that a dual class structure in and of itself negatively influences firm performance. The dual class nonfamily firm variable bears a positive coefficient estimate (β 6), although not significant at conventional levels, suggesting that dual class structures sans family ownership appear to have little effect on firm valuations. Family presence however, exhibits a negative and significant relation to Tobin s Q with a particularly large discount for the combination of dual class and family firm. The results of the matched sample largely reflect those of the full sample. As noted earlier, the vast majority of dual class firms (89%) are also family firms, thus blurring the effect of family ownership and dual class structures on firm valuation. To provide further 10 We calculate this as the coefficient estimate on family firm divided by the average Tobin s Q for the full sample; /1.96 = %. 15

17 insights into the effect of dual class structures on valuation, we develop a matched sample of dual class nonfamily firms and single class nonfamily firms. That is, we drop family firms from the analysis and create a matched sample of single- and dual- class nonfamily firms, thus allowing us to separate the family effect from the dual class effect. The sample consists of 452 firm-year observations, providing 226 single class firms and 226 dual class firms (all nonfamily). 11 After removing the family firm effect from the analysis, we find that dual class structures exhibit better valuations than their single class peers. Specifically, the coefficient estimate on dual class in column 6 of Table 4 is positive and significant at the 5% level or better, indicating that dual class firms carry valuation premiums of 27.4% versus single class firms. We calculate this premium as the coefficient estimate on dual class divided by the average Tobin s Q for this subsample (=0.549/2.000 = 27.4%). Our analysis paints a somewhat different picture from conventional wisdom on the effects of dual class structures on firm valuation. The results suggest that dual class structures without family owners exhibit a positive relation to Tobin s Q. Dual class shares only exhibit a detrimental influence on firm valuations in the presence of family shareholders, suggesting that family ownership accounts for an important covariate in estimating the effect of dual class structures on firm valuations. The Russell 3000 captures nearly 98% of total U.S. market capitalization, suggesting that our sample of dual class firms likely provides a relatively fair representation of the prevalence of dual class firms. Although we find that dual class firms without family owners exhibit a positive relation to equity valuations, we note an important qualification to our analysis in that the vast majority of dual class firms are also family firms (89%). The dual-class family-firm combination bears significant valuation discounts relative to other organizational forms. IV. The Dual Class Choice Our analysis thus far, indicates dual class structures appear to be a manifestation and continuance of founder and family ownership, and that these influential owners bear substantial valuation discounts when selling shares to outside investors. To gain a sense of the magnitude of 11 In developing this matched sample of dual- and single- class nonfamily firms, we use coarsened exact matching (CEM) with an exact match on Fama-French industry code and closest match on firm size (total assets) and firm age. All firms match on industry. The difference in match on total assets (firm age) between dual- and single class firms as a percent of average firm size (firm age) in the subsample is 9.94% (4.33%). Internet appendix II provides summary statistics for this matched sample. 16

18 the discount that dual class family owners suffer when selling shares to the investing public, we conduct a few simple calculations with our data. The average market capitalization for a dual class family firm in our sample is $3.463 billion with the family holding 30.1% of the cash flow rights and 58% of the voting rights. Assuming that a hypothetical firms goes from private ownership to a dual class family firm (overnight) with similar characteristics to our sample averages, the family would leave about $324 million on the table in the offering. 12 Dual class family owners maintain their stakes, on average, for about 54 years, potentially allowing ample time for these influential owners to extract the loss from outside shareholders through private benefits of control (e.g., prerequisites, excess compensation, nepotism, empire building, etc.) (Gompers et al., 2010). Yet, extant literature indicates that family owners can control the firm with far less than 58% of the firm s voting power and still extract substantial private benefits. Our own analysis further indicates that that these influential owners can leave far less on the table by establishing single class structures. The earlier univariate analysis indicates that dual class structures cluster in a relatively small number of industries with 10 of the Fama-French industry groups accounting for nearly 65% of dual class firms; suggesting that factors beyond moral hazard influence the dual class choice. We undertake an exploratory investigation and ask whether the industrial or business environment affects family owners decision to establish super voting share classes. Our analysis uses a set of industry structure variables that potentially relate to families decision to retain an ownership stake and dual class shares. As large, concentrated shareholders with the majority of their wealth tied to a single firm (average dual class family stake is $1.04 billion), family owners arguably possess strong incentives to protect their investment. We use the following specification to examine the relation between the business environment and equity ownership structure; Equity Structure = α + β(industry Characteristics) + X β + ε (2) Where; Equity structure = binary variables for single class nonfamily firms, dual class nonfamily firms, single class family firms, and dual class family firms, i.e., four dependent variables. Industry characteristics = growth opportunities, sales growth, innovation, branding, asset (in)tangibility, asset intensity, and profitability. All measured at the Fama-French 48 industry level, excluding dual-class firms. 12 We calculate this as: (Avg. Q/(Avg. Q * (1 Dual class family discount)) * Mkt cap dual class family * 0.70) (Mkt Cap of Dual Class Family *0.70) = (1.96/(1.96 * ( ))*3,463.57*0.70) (3,463.7 * 0.70) = $ million. 17

19 Firm-level control variables size (natural log of total assets), firm age (natural log years since inception), firm leverage (long term debt to assets), firm performance (return on assets), firm risk (standard deviation of stock returns for the prior 36 months), and innovation (R&D-to-sales). We have four dependent variables in our specification, covering the four possible combinations between dual and single class firms, and family and nonfamily firms. To allow for a statistical comparison of industry traits on the four different equity structure choices, we use multinomial logistic regressions. 13 With this technique, we choose a base case (single class nonfamily firms) and then compare the other three cases (dual class nonfamily, single class family, and dual class family) against the base case in a system of simultaneous logistic regressions. Thus, our system of equations is; Eq. 1: Dual class nonfamily to single class nonfamily = α + β 1(Industry Characteristics) + Xβ + ε Eq. 2: Single class family to single class nonfamily = α + β 2(Industry Characteristics) + Xβ + ε Eq. 3: Dual class family to single class nonfamily = α + β 3(Industry Characteristics) + Xβ + ε The coefficient estimates from equation 1 compares dual class nonfamily firms to single class nonfamily firms (base case). Similarly, equation 2 and equation 3 compare single class family firms and dual class family firms, respectively, to the base case. Because we estimate our regressions in a system of equations, we can then conduct Wald χ 2 -tests to examine the equality of coefficient estimates between the equations (e.g., β 1 = β 2, from Eq. 1 and Eq. 2). Table 5, Panels A and B, presents the results of the multinomial logistic regressions. We exclude dual-class firms when calculating the industry measures to remove confounding effects, if any, of dual class firms on aggregate industry averages. 14 Panel A investigates three measures of industry growth relative to equity structure. These are; industry growth opportunities (average Tobin s for firms in each industry), industry sales growth (five year rolling industry sales growth), and industry innovation (average industry R&D to sales). To simplify interpretation of the multinomial logit regressions, we develop binary variables that equal one for the top quartile of 13 A straightforward method to obtain consistent and asymptotically efficient coefficient estimates would be to run individual logistic regressions for each equity structure choice (e.g., dual class family firm) relative to each industry trait (e.g., industry growth rate). Individual logistic regressions however, create a drawback because we cannot statistically compare coefficient estimates between the different regression specifications and thus, cannot make economic inferences as to the effect of industry traits on the four different equity structures. Notably, the coefficient estimates from the individual logistics regressions and multinomial logistics are nearly identical in magnitude and significance. 14 When including dual class firms in the industry averages, we find coefficient estimates and significance levels nearly identical to those when excluding dual class firms from the industry averages. 18